Embarrassment guides language choice

English is the language in which I best express the subtlety of life. But I must say that French is the language closest to my heart. And for this same reason, English gives me a sufficient distance to write.

– Yann Martel (Author of Life of Pi)

Talking about taboos is aversive, but languages around the world provide a way to alleviate the pain. They borrow words from foreign languages for embarrassing topics such as sex (Burridge & Benczes, 2019). English speakers use “derrière” and “ménage à trois” from French, Cantonese speakers use English-based euphemisms to refer to condoms (“dom dom”), and colloquial Hebrew borrows vulgar curses from Arabic. Here we study a more general phenomenon: Bilinguals may shift into a foreign language altogether to reduce embarrassment.

We argue that the anticipation of embarrassment could guide bilinguals’ language choice. Bilinguals typically have a native language, also often referred to as a first language, that is acquired first, spoken most proficiently, and learned naturally within their community. Late bilinguals also have at least one foreign language, also referred to as a second language, that is acquired later, primarily through classroom learning (Dewaele, 2010). While both labels are used in multilingual research, we use “native” and “foreign” to focus on the context of acquisition rather than the order of learning (Caldwell-Harris, 2014).

Language can influence both the emotional resonance of retrieved memories and the experience of emotions. For instance, replacing first-person pronouns (e.g., “me,” “I”) with third-person pronouns (e.g., “he,” “she”) has been shown to reduce negative affect during memory retrieval by promoting a more distanced perspective (Kross et al., 2014). This language-induced distance extends to bilingualism as well, as bilinguals typically learn foreign languages in formal settings where it is difficult to integrate affective experiences with words (Pavlenko, 2008).

As a result, emotions tend to be experienced less automatically and intensely in a foreign tongue that is typically learned later in life (Caldwell-Harris, 2014; Harris, Ayçiçegi, & Gleason, 2003; Harris, Gleason, & Ayçiçegi, 2006; Pavlenko, 2012; Puntoni, De Langhe, & Van Osselaer, 2009; Thoma & Baum, 2019). For instance, bilinguals display heightened electrodermal reactivity to emotion-laden words when presented in their first language compared to their second language (Pavlenko, 2012). These findings are consistent with episodic trace theories, which suggest that the relationship between language and emotionality is modulated by psychological distance through context and experience (Caldwell-Harris, 2014; Pavlenko, 2008; Satpute, Nook, & Cakar, 2020; Shin & Kim, 2017). Such emotional distance afforded by a foreign language could support emotional regulation.1

In fact, researchers have found that emotional regulation strategies have different consequences depending on the language engaged in (Quiñones-Camacho, Shih, Savage, Lamar Prieto, & Davis, 2019; Thoma, 2021). For instance, Thoma (2021) found that a distraction was more effective in reducing bilinguals' negative emotions in their second language than in their first language. The reduced automaticity of affective processing may even lead to a second language advantage for some emotional regulation strategies (Morawetz, Oganian, Schlickeiser, Jacobs, & Heekeren, 2017).

Much of the existing research focuses on how people behave differently when they use their native or foreign language. Our research focuses instead on language preference and asks, what role does the anticipated emotion play in influencing language choice? We propose that people might prefer to use a foreign language to reduce anticipated aversive reactions, as a native language tends to carry stronger emotional consequences than a foreign tongue.

The literature provides only limited support for this. While there are some self-report studies and anecdotal reports, there is no experimental evidence showing that bilinguals actually prefer to use their second language to express taboo or aversive ideas.

From an emotional regulation perspective, bilinguals may choose a language to influence their emotional experience, switching into their native language to express more intense emotions (e.g., up-regulate) or into a foreign language to express less intense emotions (e.g., down-regulate) (Gross, 2015; Pavlenko, 2012). In self-report studies, multilinguals report that they express endearment to their children in their native tongue because it feels more authentic and promotes closeness (Pavlenko, 2005) and that they express anger in their native tongue because of its superior emotional force (Dewaele, 2006).

However, while self-reports reflect people's impressions of their behavior, they do not provide evidence for what language they would actually choose when expressing themselves. Indeed, self-reported responses on verbal scales often fail to show the same consistent patterns of reduced emotional reactivity in a foreign language that are found with neurophysiological or cognitive measures (Thoma, 2024). Differences in language-dependent emotional experiences may be concentrated in automatic processing, and overlooked in more conscious meta-cognitive, or anecdotal self-reports (Thoma, 2024). Thus, the present study uses cognitive and choice measures to experimentally investigate the idea that bilinguals' language preferences are guided by anticipated emotions.

A language preference is also alluded to in clinical settings, where therapists sometimes observe bilingual patients switching to their less dominant language, where emotion words are less richly represented, to express feelings of anger, loss, and bereavement (Santiago-Rivera, Altarriba, Poll, Gonzalez-Miller, & Cragun, 2009). Other accounts note bilingual clients switching to their foreign language to talk about traumatic and threatening experiences, presumably to distance themselves from difficult memories (Altarriba & Santiago-Rivera, 1994; Javier, 1989; Krapf, 1955; Movahedi, 1996; Santiago-Rivera et al., 2009). These accounts point towards the idea that bilinguals may strategically use their languages to modulate their emotional experiences.

Such anecdotal reports align with our theory, but they are inherently biased because the emotion-guided language choice that they demonstrate is a "one-sided event" (Gilovich, 2008). An event is one-sided when its presence is noticed but its absence goes unnoticed, so people have no opportunity to refute their beliefs because they do not encode the non-events. This is a fundamental problem with anecdotes and memory-based reports. When a psychotherapy patient shifts into a foreign language to talk about trauma, the event is noticed because the patient switched languages. When the patient continues to talk about trauma in their native tongue, the therapist is unlikely to notice it, because the act of not switching languages is a non-event. Due to this asymmetry, a relationship between the traumatic nature of the topic and language switching cannot be established. Therefore, such observations do not provide reliable evidence for the idea that anticipated emotional reaction leads to the language shift.

Thus, although related ideas have been explored, there is no experimental evidence supporting the claim that anticipated emotions guide bilinguals' language preferences or that such preferences function as a form of emotional regulation. Prior research has primarily examined language behavior in response to emotional experiences rather than in anticipation of them (Pavlenko, 2012). For example, in an observational study, Williams, Srinivasan, Liu, Lee, and Zhou (2020) found that bilinguals' expression of negative emotions predicted their code-switching behavior, and they attributed this to emotional arousal interfering with the cognitive control processes that help maintain the use of a particular language. However, such findings concerning rapid-code switches may not apply to language choice, which involves sustaining a language over a longer period of time. Moreover, the bidirectional nature of such switches challenges the idea that bilinguals use their second or foreign language to create emotional distance. These findings do not provide evidence of a bilingual language preference that reflects a proactive emotional regulation strategy.

Only therapy accounts make a similar claim that bilinguals switch to their foreign language to create distance from negative topics (Santiago-Rivera et al., 2009). However, such claims are anecdotal, and code-switching often depends on language availability, including the therapist's ability to use either language.

Here we report the first experimental evidence that language choice is guided by anticipated emotion, focusing on the impact of anticipated embarrassment. In these studies, participants chose between a native and foreign language in anticipation of speaking publicly about a range of embarrassing topics, including disgusting bodily functions (Studies 1, 2a, 2b, and 3), socially sensitive revelations (Studies 1, 2a, and 2b), and sexual activity (Study 4a and 4b). We evaluated our proposal in several languages, including Chinese, English, and Spanish.

While we argue that language choice in our experiments is guided by the difference in anticipated embarrassment between the native and foreign language, we also evaluated a competing account that suggests that our findings are a function of cultural elements. It is possible that the language choice is guided by cultural considerations that are associated with each language, rather than the native-ness of the language. To evaluate this, we crossed two languages by conducting the same experiment with native English speakers using Spanish as the foreign language (Studies 2a and 4a) and native Spanish speakers using English as the foreign language (Studies 2b and 4b). We also investigated emotional and social distancing as potential mechanisms by evaluating participants' expected emotional and social consequences for discussing embarrassing content in a native and foreign language. Lastly, we evaluate language choice using a more naturalistic approach with Spanish-English bilinguals who are native English speakers (Study 4a) or native Spanish speakers (Study 4b).

All de-identified data and materials are available on Open Science Framework (https://osf.io/gsyb5/?view_only=b0cf4de7087143d3aa644f16305356a2). Studies 2a, 2b, 4a, and 4b were preregistered on AsPredicted (Study 2a: http://aspredicted.org/blind.php?x=bu3yt8, Study 2b: https://aspredicted.org/blind.php?x=z8qe4m, Study 4a: https://aspredicted.org/xgwz-g2bk.pdf, Study 4b: https://aspredicted.org/rhbj-ndkk.pdf).

1. General method

To avoid redundancy, we report here elements of the method shared across studies. In all studies, participants were presented with embarrassing and emotionally neutral topics and were told they would read these topics aloud to an audience or be recorded. They were then asked to choose whether to say each topic in their native or foreign language. Participants did not actually read anything aloud during the experiment, but to ensure that the language choice felt real, this was only disclosed after participants indicated their language choices. Instructions were presented in the participants' native language, and the statements appeared in both languages, side-by-side, so that participants could imagine the experience of saying them in either language. The statements were translated and back-translated to ensure they were comparable across languages (Brislin, 1970). Our research was approved by a university Institutional Review Board and followed relevant ethical regulations, including obtaining informed consent from all participants. In order to participate in the studies, participants had to report knowing their foreign language at an intermediate level or above, pass a short reading comprehension test in their foreign language, grow up not speaking their foreign language at home, consider their native language to be their dominant language, and their self-reported foreign language proficiency could not be higher than their native language proficiency. Participants were also required to pass bot detections and report not participating in any previous studies on the same topic. The Supplementary Materials contain participant characteristics, stimuli from all studies, as well as some additional analyses.

2. Study 1

In our first test of the hypothesis, we recruited native Mandarin speakers who speak English as a foreign language to examine their language choices for embarrassing and emotionally neutral statements.

We designed the stimuli in a way that minimized the confounding effect of cultural norms and self-expression. While norms allow explicit, direct communication in some cultures, others involve the use of implicit, indirect manners as a form of self-expression (Lee, Shrum, & Yi, 2017). It is possible, then, that the choice of language can be the result of cultural assumptions about cultures that use English. Therefore, we chose statements where cultural norms for self-expression should not matter.

2.1. Method

2.1.1. Participants

A power analysis indicated that a minimum of 208 participants were required to achieve a minimum power of 0.80 for a two-tailed one-sample Wilcoxon signed-rank test, assuming a small effect size (d = 0.2). To get a robust outcome, we planned for a larger sample of 400. We screened participants' language background before the study using the criteria already described (see Table S1, Supplementary Materials for Eligibility Criteria Summary), and participants were deemed ineligible if they fell outside of the eligibility criteria. We recruited native Mandarin speakers who spoke English as a foreign language to participate in an in-person experiment at a university center in Beijing. Six participants had technical difficulties and were replaced, resulting in a final sample of 402 participants.2 69 % were female, and the mean age at which they had begun learning English was 9.7. The mean English proficiency score was 5.07 (SD = 1.17), and the mean Mandarin score was 6.59 (SD = 0.72) (see Table S15, Supplementary Materials for Language Proficiency Breakdown).

2.2. Procedure

Participants were told they would participate in a two-part study, the first part to be conducted that day and the second session on another day. In reality, the experiment only consisted of one part, and participants were thanked and compensated a few days later for both sessions. Participants were told the goal was to understand how people communicate and that they would need to read sentences out loud in front of other people, including the researcher and possibly other participants in the second session. As part of session one, they were asked to review the sentences they would read aloud in session two and to indicate the language they would use for each sentence. Participants were presented with sentences where they had to first fill in a blank reflecting their experience and that best described themselves, presumably to read it out loud. For example, they received the sentence "I don't always ___ after I pee" with the following options "wash my hands; wipe my privates; flush the toilet." Each pair of sentences had one embarrassing and one neutral statement, and participants had to choose which statement to say in Mandarin and which in English. Participants completed one practice round before they reviewed the actual items.

Thirty pairs of statements were divided into disgust and intrusive sets. Embarrassing statements from the disgust set asked participants to describe disgusting bodily functions and habits, such as body parts they like to scratch, the shape of their excrement, and so on. Embarrassing statements from the intrusive set asked participants to divulge sensitive and private information. For example, they were asked to talk about a time when they revealed information that a friend confided in them, thereby violating the friend's trust or a time when they lied about their qualifications. The neutral statements matched the embarrassing statements in length and difficulty of vocabulary.

After indicating their language choices, participants were randomly assigned to imagine reading the disgust statements and the intrusive statements out loud, in counterbalanced order. We then evaluated their expectations of the consequences of using each language to speak about the embarrassing statements. While the embarrassing statements were presented in both Chinese and English so that participants could imagine the experience of reading them out loud in either language, the following four questions were presented in Chinese, the participants' native language.

1. Discomfort: To what extent would reading aloud these sentences in Chinese/English make you feel uncomfortable? (1 = Not at all uncomfortable, 7 = Extremely uncomfortable)
2. Negative social judgment: When reading aloud these sentences in Chinese/English, how negatively do you think others would evaluate your character? (1 = Not at all negative, 7 = Extremely negative)
3. Reveal kind of person: When reading aloud these sentences in Chinese/English, how much could what you say reveal what kind of person you are? (1 = Reveals nothing, 7 = Reveals a lot)
4. Reveal personality: Do you think others would feel that the answers you would give in Chinese/English would reflect your true personality? (1 = Not at all, 7 = Very much so)

Participants also reported the extent to which the languages make them feel more emotional (1 = More emotional when speaking in English, 4 = No difference between the languages, 7 = More emotional when speaking in Chinese) and the extent to which the languages make them feel more like themselves (1 = More like myself when speaking in English, 4 = No difference between the languages, 7 = More like myself when speaking in Chinese).

Finally, participants reported their expectations about native Chinese and native English speakers' tendency to disclose information about themselves ("What would a native Chinese/English speaker think?" 1 = Strongly disagree, 7 = Strongly agree).

1. Brevity: When I talk about my feelings, what I say is usually very brief.
2. Revealing: I frequently reveal private things without hesitation.
3. Opinions: I often express my personal views and opinions.

2.3. Results and discussion

Recall that participants received pairs of statements (embarrassing/neutral) and had to decide in which language to talk about each statement (native/foreign). If language choice is independent of the statement's level of embarrassment, then it should be at chance (50 %).

Participants chose English, a foreign language, to read aloud the embarrassing statements in 78 % of the statement pairs, which was significantly higher than chance, $Z = -15.05$, $p < .001$, $r = 0.75$ (See Fig. 1). Participants chose the foreign language more frequently for the disgust statements compared to the intrusive statements, paired $t(401) = 10.24$, $p < .001$, $d_{\text{Cohen}} = 0.51$. For the disgusting statements, participants chose English to read aloud the embarrassing statements in 85 % of the statement pairs, and this was significantly higher than chance, one-sample $t(401) = 28.48$, $p < .001$, $d_{\text{Cohen}} = 1.42$. For the intrusive statement pairs, participants chose English for the embarrassing statement 72 % of the time, and this was significantly higher than chance, one-sample $t(401) = 15.2$, $p < .001$, $d_{\text{Cohen}} = 0.76$. Importantly, there are no established differences between Chinese and English-speaking cultures regarding the conversational acceptability of disgusting and intrusive statements that could explain the results. Therefore, it is unlikely that such cultural considerations could account for our findings.

Participants expected significantly less severe emotional and social costs for reading aloud the embarrassing statements in English compared to Mandarin. Specifically, they anticipated that personal discomfort and negative social judgments would be attenuated, and they anticipated that their statements in English would reflect less significantly upon their character and personality (Table 1).

These expectations may contribute to the language preferences. Participants were more likely to choose English for the embarrassing statements if they anticipated more severe social and emotional costs in their native language. Measures of discomfort, negative social judgment, and revealing their personality predicted language choice for both disgust and intrusive statements, while the measure of revealing the kind of person they are, significantly predicted language choice only for the intrusive statements ($p < 0.05$). Among these measures, language choice for both types of statements was most strongly associated with discomfort ($r = 0.45$ and $r = 0.33$ for disgust and intrusive statements respectively). These results suggest that language could serve as a distancing or emotional regulation strategy. Participants chose a foreign language, which they thought would reduce the anticipated emotional and social costs compared to a native language, to talk about embarrassing statements. Section 7 in the Supplementary Materials contains further analyses.

Participants also indicated that Mandarin made them feel more like themselves than English, one-sample $t(401) = 11.55$, $p < .001$, $d_{\text{Cohen}} = 0.58$, and more emotional, one-sample $t(401) = 12.40$, $p < .001$, $d_{\text{Cohen}} = 0.62$. Compared to the experience of using a native language, foreign languages felt less emotionally involved and less self-relevant, and such expectations could drive language choice.

Participants expected that native English speakers would be more likely to be brief when talking about their feelings, reveal private information without hesitation, and express their personal opinions (Table 2). A greater expectation of native English speakers to reveal private information and express their opinions compared to native Chinese speakers was associated with a greater likelihood of choosing to say embarrassing intrusive statements aloud in English. However, this association was only observed for intrusive items and did not extend to disgust-related items. Thus, cultural expectations of self-disclosure among Chinese and English speakers do not fully account for participants' language choices.

We used R (R Core Team, 2022) and lme4 (Bates, Maechler, Bolker, & Walker, 2015) to perform a linear mixed-effects analysis of the relationship between language choice and statement. Given that each statement pair had both a neutral and an embarrassing item, we specified only the language choice for embarrassing items and estimated the model including item type (intrusive vs. disgust), proficiency, and the interaction between item type and proficiency as fixed effects, and participant-level and item-level random effects. The intercept value, reflecting the likelihood of participants choosing to use their foreign language for embarrassing topics compared to neutral topics, remained a significant predictor ($\beta = 2.17$, $p < .001$) when controlling for item type ($\beta = -0.48$, $p < .001$), proficiency ($\beta = 0.05$, $p = .018$), and the interaction variable ($\beta = 0.05$, $p = .02$), as well as participant-level, and item-level random effects. This suggests that the main effect of selecting a foreign language for embarrassing topics is not driven by item type, proficiency, or their interaction.

3. Studies 2a and 2b

A cultural account for our results suggests that a language context could trigger norms that are culturally relevant and influence participants' language choices, independent of whether the language is native or foreign. Study 1 addressed this possibility by using statements that are unlikely to be linked to a particular culture or its norms. It is still possible, however, that the effect could be the result of cultural norms associated with a given language. Studies 2a and 2b provide a direct and stronger test of this account by crossing the languages.

Study 2a examined the phenomenon with native English speakers who speak Spanish as a foreign language, and Study 2b did so with native Spanish speakers who speak English as a foreign language. A cultural account for our results would suggest that a given language would be preferred for a particular statement because of its role in the culture; therefore, people would say that statement in the language that is associated with the culture. This predicts that language choice would be independent of language native-ness. In contrast, if language choice for embarrassing statements depends on whether the language is native or foreign, then native-ness should determine language choice in Studies 2a and 2b.

3.1. Method

3.1.1. Participants

Study 2a (pre-registered). Since the effect size of Study 1 was larger than our original power analysis assumed ($d = 0.2$), we conducted a second analysis. This resulted in 94 participants ($d = 0.3$) for a two-tailed one-sample Wilcoxon signed-rank test with a minimum power of 0.80. Aiming for a robust outcome, we recruited 110 participants for Study 2a and the same for 2b. Three participants were replaced because of failed pre-screen attempts and inconsistent answers. The final sample consisted of 111 native English speakers who speak Spanish as a foreign language (65 % female, $M_{\text{age of acquisition}} = 12.86$). The mean Spanish proficiency score was 4.94 (SD = 0.86), and the mean English score was 6.99 (SD = 0.04).

Study 2b (pre-registered). 110 native Spanish speakers from Puerto Rico3 who speak English as a second language participated in an online study. There were no exclusions or replacements (70 % female, $M_{\text{age of acquisition}} = 6.26$). The mean English proficiency score was 6.63 (SD = 0.63), and the mean Spanish score was 6.98 (SD = 0.13).4 In addition to meeting the eligibility criteria previously described, participants also had to be born and/or raised in Puerto Rico.

3.2. Procedure

Both experiments followed the procedure of Study 1 and used the same statements with minor adjustments for cultural appropriateness. To make sure participants truly anticipated speaking to an audience, they were told that they would first make their language choices in an online survey, and then participate in a videoconference (Study 2a, 2b) or an in-person study (Study 2a), where they would verbally respond to the questions in front of the experimenter and other participants. After making language choices for all statements, participants described the criteria that informed their choices in a free response format. After data collection, two native English-speaking assistants (Study 2a) and two native Spanish-speaking assistants (Study 2b) coded the explanations. Participants then rated the emotional and social consequences they anticipated if they were to read aloud both disgust and intrusive statements in English and Spanish, using the same four measures from Study 1. Different from Study 1, participants rated the anticipated consequences for all statements in a repeated measures design. Finally, participants reported whether they felt more emotional and more like themselves when speaking in English or Spanish. In Study 2a, participants completed the same disclosure scale used in Study 1 whereas in Study 2b, participants answered a shortened version that included only the revealing item of the scale.

3.3. Results and discussion

Participants preferred to use their foreign language to speak about embarrassing statements. In Study 2a, native English speakers chose Spanish, the foreign language, to read aloud the embarrassing statements in 58 % of the statement pairs, which was significantly higher than chance, $Z = -3.04$, $p = .002$, $r = 0.29$ (See Fig. 1). For disgust statements, participants chose a foreign language for the embarrassing statement in 66 % of the statement pairs. This was significantly higher than chance, $t(110) = 6.15$, $p < .001$, $d_{\text{Cohen}} = 0.58$. For the intrusive statement pairs, participants chose a foreign language for the embarrassing statement only 50 % of the time.

In Study 2b, native Spanish speakers chose English, their foreign language, to read aloud the embarrassing statements in 63 % of the statement pairs, which was significantly higher than chance, $Z = -6.18$, $p < .001$, $r = 0.59$ (See Fig. 1). For disgust statements, participants chose English for the embarrassing statement in 68 % of the statement pairs, and this was significantly higher than chance, $t(109) = 8.90$, $p < .001$, $d_{\text{Cohen}} = 0.85$. For the intrusive statement pairs, participants chose English for the embarrassing statement 59 % of the time, and this was also significantly higher than chance, $t(109) = 4.22$, $p < .001$, $d_{\text{Cohen}} = 0.40$. In both Study 2a and Study 2b, participants' choice for using a foreign language was stronger for the disgust statement compared to the intrusive statements, paired $t(110) = 7.35$, $p < .001$, $d_{\text{Cohen}} = 0.70$ and paired $t(109) = 5.22$, $p < .001$, $d_{\text{Cohen}} = 0.50$, respectively.

Just like in Study 1, participants in both Study 2a and 2b expected that using a foreign language would increase emotional and interpersonal distance. In Study 2a, they indicated that using Spanish made them feel less like themselves, one-sample $t(110) = -18.12, p < .001, d_{\text{Cohen}} = 1.72$, and less emotional, one-sample $t(110) = -6.97, p < .001, d_{\text{Cohen}} = 0.66$. In Study 2b, participants reported that English made them feel less like themselves, one-sample $t(109) = 8.45, p < .001, d_{\text{Cohen}} = 0.81$, and less emotional, $t(109) = 7.98, p < .001, d_{\text{Cohen}} = 0.76$. Furthermore, participants in both studies expected that discussing embarrassing statements in the foreign language would carry less serious consequences. Compared to using their native tongues, participants anticipated that using a foreign language would be less uncomfortable and generate less negative social judgment. They also imagined that what they said in the foreign language would be less reflective of who they were and their true personality (Tables 3 and 4). It is especially noteworthy that Studies 2a and 2b showed the same pattern of results even when the languages were crossed. The fact that language choice was a function of the native-ness of the language, not of the language itself, supports the native-ness account and rules out the cultural association account.

Again, these expectations were associated with the language choice for taboo statements. In both studies, participants were more likely to choose a foreign language for embarrassing statements when they anticipated more severe emotional and social consequences in their native language. In both Study 2a and 2b, language choice for disgusting statements most strongly correlated with anticipated discomfort (2a: $r = 0.25, 2b: r = 0.37$) and negative social judgments (2a: $r = 0.23, 2b: r = 0.36$), while language choice for intrusive statements most strongly correlated with the extent to which they believed their answers would reveal their personality (2a: $r = 0.33, 2b: r = 0.36$). These results suggest that the emotional and interpersonal distance a foreign language is expected to provide for aversive statements may play a role in language choice. Participants chose a foreign language, which provided distance from the expected emotional and social costs of speaking about aversive statements. Section 7 in the Supplementary Materials contains further analyses for the measures of anticipated consequences.

Participants in Study 2a expected that native Spanish speakers would be less likely to be brief when talking about their feelings (Table 5), and participants in Study 2b expected native English speakers to more frequently reveal private and personal matters (Table 6). However, in both Studies 2a and 2b, cultural expectations did not correlate with participants' language choice in either the disgust set or the intrusive set. Thus, participants' language choices cannot be explained by cultural expectations of self-disclosure among native English or Spanish speakers.

We then leveraged the counterbalanced nature of Studies 2a and 2b by collapsing them into a single analysis to further examine the preference of a foreign language for speaking about embarrassing statements (See Table 7). Participants chose a foreign language to read aloud embarrassing statements in 61 % of the statement pairs, significantly above chance, $Z = -6.53, p < .001, r = 0.45$. This preference was also present for disgust statements (67 %), $t(220) = 10.30, p < .001, d_{\text{Cohen}} = 0.70$, and intrusive statements (54 %), $t(220) = 2.45, p = .01, d_{\text{Cohen}} = 0.17$. Supporting our previous finding, we find that participants' choice for using a foreign language was stronger for the disgust items compared to the intrusive items, paired $t(220) = 8.89, p < .001, d_{\text{Cohen}} = 0.60$.

Finally, participants indicated that using a foreign language made them feel less emotional, one-sample $t(220) = 10.58, p < .001, d_{\text{Cohen}} = 0.71$, and less like themselves, one-sample $t(220) = 16.99, p < .001, d_{\text{Cohen}} = 1.14$.

The qualitative analyses show that participants in both Studies 2a and 2b chose a foreign language for the embarrassing statements beyond concerns about limited vocabulary or language proficiency. Instead, language choices correlated with concerns about aversive emotions. In Study 2a, 82 % of participants mentioned concepts related to embarrassment or negative emotion as informing their language choices (Cohen's $\kappa = 0.82, z = 8.10, p < .001$), while 43 % mentioned concepts related to vocabulary (Cohen's $\kappa = 0.87, z = 8.68, p < .001$). Participants who reported using embarrassment as a criterion for language choice were more likely to choose the foreign language, Spanish, for the embarrassing statements (62 %) compared to those who did not mention embarrassment (39 %), $\beta = 0.64, SE = 0.17, \chi^2(1) = 14.10, p < .001, OR = 1.90$. On the other hand, participants who mentioned using vocabulary as a criterion were less likely to choose Spanish for the embarrassing statements (47 %) compared to those who did not mention vocabulary (66 %), $\beta = -0.53, SE = 0.13, \chi^2(1) = 15.88, p < .001, OR = 0.59$.

In Study 2b, 51 % of participants referred to concepts related to embarrassment or negative emotion as having guided their language choices (Cohen's $\kappa = 0.84, z = 8.47, p < .001$), while 48 % referenced concepts related to vocabulary (Cohen's $\kappa = 0.72, z = 7.19, p < .001$). Participants who reported using embarrassment as a determinant of language choice were more likely to choose English, the foreign language, for the embarrassing statements (67 %) compared to those who did not mention embarrassment (57 %), $\beta = 0.28, SE = 0.09, \chi^2(1) = 8.78, p < .01, OR = 1.32$. Participants who mentioned using vocabulary as a determinant of language choice were just as likely to choose English for the embarrassing statements (62 %) as those who did not mention vocabulary (63 %), $\beta = -0.03, SE = 0.10, \chi^2(1) = 0.11, p = .74, OR = 0.97$, suggesting that vocabulary concerns could not have been responsible for the observed pattern of language choice.

Even after crossing the languages, people still chose their respective foreign languages to speak about taboo topics, suggesting the phenomenon cannot be explained by culture.

We next consider an alternative explanation for our results. It is possible that participants might have zero or limited experience with certain embarrassing statements. This might have led participants to feel like they were lying, and this may have induced the language effect instead of the anticipated embarrassment. To examine this account, we evaluated the 'common-ness' of each statement with a separate sample of participants using the same inclusion criteria of Study 2a. We found that, on average, embarrassing items ($M = 2.50, SD = 0.77$) were indeed rated as less common than their matched controls ($M = 3.43, SD = 0.84$), $t(29) = -4.27, p < .001, d_{\text{Cohen}} = 1.15$. However, this difference in experience between embarrassment and neutral statements did not correlate with language choice ($r = -0.12, p = .50$). This suggests that experience with a statement was not responsible for language choice in this study. Yet, because the common-ness ratings were obtained with a different group of participants, it might not be shared with participants from the original study. To evaluate whether experience with an embarrassing topic drives the language effect, we conducted Study 3.

Measure	Disgust			Intrusive
	Native	Foreign	Compare	Native	Foreign	Compare
Discomfort	6.18 (1.28)	4.44 (1.57)	$t(200) = 15.94, p < .001, d = 1.12$	5.54 (1.53)	4.32 (1.55)	$t(200) = 12.61, p < .001, d = 0.89$
Negative Social Judgment	4.51 (1.33)	3.83 (1.52)	$t(200) = 13.79, p < .001, d = 0.97$	4.71 (1.54)	3.84 (1.55)	$t(200) = 8.87, p < .001, d = 0.63$
Reveal Kind of Person	5.80 (1.92)	4.51 (1.61)	$t(200) = 5.95, p < .001, d = 0.42$	5.45 (1.80)	4.55 (1.52)	$t(200) = 7.48, p < .001, d = 0.53$
Reveal Personality	5.00 (1.61)	4.02 (1.53)	$t(200) = 8.60, p < .001, d = 0.61$	5.01 (1.68)	4.01 (1.53)	$t(200) = 8.53, p < .001, d = 0.60$

Measure	Native Chinese Speakers	Native English Speakers	Compare	Correlations with Language Choice for Disgust Items	Correlations with Language Choice for Intrusive Items
Brevity	3.79 (1.53)	4.04 (1.36)	$t(400) = -2.34, p = .02, d = 0.12$	$r(399) = -0.05, p = .32$	$r(399) = -0.06, p = .20$
Revealing	2.61 (1.56)	2.85 (1.61)	$t(401) = -2.14, p = .03, d = 0.11$	$r(400) = -0.01, p = .88$	$r(400) = -0.12, p = .02$
Opinions	3.81 (1.35)	5.86 (1.07)	$t(401) = -23.09, p < .001, d = 1.15$	$r(400) = -0.09, p = .08$	$r(400) = -0.12, p = .01$

Measure	Disgust			Intrusive
	Native	Foreign	Compare	Native	Foreign	Compare
Discomfort	5.49 (1.52)	4.23 (1.75)	$t(110) = 8.25, p < .001, d = 0.78$	4.77 (1.65)	4.04 (1.58)	$t(110) = 4.94, p < .001, d = 0.47$
Negative Social Judgment	5.00 (1.65)	4.23 (1.58)	$t(110) = 6.11, p < .001, d = 0.58$	4.95 (1.75)	4.32 (1.59)	$t(110) = 4.34, p < .001, d = 0.41$
Reveal Kind of Person	4.05 (1.81)	3.40 (1.59)	$t(110) = 4.69, p < .001, d = 0.45$	4.41 (1.77)	3.78 (1.62)	$t(110) = 4.29, p < .001, d = 0.41$
Reveal Personality	4.14 (1.91)	3.18 (1.67)	$t(110) = 6.51, p < .001, d = 0.62$	4.41 (1.88)	3.49 (1.65)	$t(110) = 5.98, p < .001, d = 0.57$

Measure	Disgust			Intrusive
	Native	Foreign	Compare	Native	Foreign	Compare
Discomfort	4.18 (2.21)	3.16 (1.84)	$t(109) = 5.83, p < .001, d = 0.56$	3.77 (2.19)	2.83 (1.79)	$t(109) = 5.70, p < .001, d = 0.54$
Negative Social Judgment	4.13 (2.08)	3.63 (1.76)	$t(109) = 3.45, p < .001, d = 0.33$	4.33 (2.18)	3.69 (1.95)	$t(109) = 4.15, p < .001, d = 0.39$
Reveal Kind of Person	3.59 (1.91)	3.28 (1.76)	$t(109) = 2.18, p < .05, d = 0.21$	3.72 (1.96)	3.31 (1.90)	$t(109) = 2.98, p < .01, d = 0.28$
Reveal Personality	3.60 (2.09)	3.03 (1.73)	$t(109) = 3.22, p < .01, d = 0.31$	3.80 (2.09)	3.26 (1.88)	$t(109) = 3.24, p < .01, d = 0.31$

Measure	Native English Speakers	Native Spanish Speakers	Compare	Correlations with Language Choice for Disgust Items	Correlations with Language Choice for Intrusive Items
Brevity	4.78 (1.30)	3.50 (1.49)	$t(110) = -6.75, p < .001, d = 0.64$	$r(109) = -0.17, p = .08$	$r(109) = -0.03, p = .79$
Revealing	3.76 (1.31)	3.37 (1.51)	$t(110) = -1.95, p = .053, d = 0.18$	$r(109) = -0.05, p = .61$	$r(109) = -0.09, p = .35$
Opinions	4.75 (1.23)	5.02 (1.31)	$t(110) = 1.64, p = .10, d = 0.16$	$r(109) = 0.10, p = .30$	$r(109) = -0.06, p = .55$

Measure	Native Spanish Speakers	Native English Speakers	Compare	Correlations with Language Choice for Disgust Items	Correlations with Language Choice for Intrusive Items
Revealing	4.04 (1.80)	3.56 (1.58)	$t(109) = 2.60, p = .01, d = 0.25$	$r(108) = -0.02, p = .84$	$r(108) = -0.07, p = .48$

4. Study 3

Studies 1, 2a, and 2b show that participants preferred to use their foreign languages to talk about taboo or embarrassing statements. Study 3 addresses the possibility that this may be due to less familiarity with the taboo statements. One might be concerned that a lack of experience with some topics led participants to feel like they were lying when they made a statement about them. It is possible that they preferred to lie in a foreign language or felt more comfortable doing so because of the psychological distance it provides (but see Bereby-Meyer et al., 2020 for the opposite). If this is the case, then language choice in our studies might not reflect anticipated embarrassment. To evaluate this possibility, Study 3 followed a similar procedure and methodology as the previous studies, while examining whether experience with the statements contributed to people's language choice. It is also possible that participants' language preferences align with cultural norms associated with native English or Spanish speakers. If so, preferences for discussing embarrassing topics should be driven by cultural expectations regarding openness about personal information.

4.1. Method

4.1.1. Participants

Following the sample size from Studies 2a and 2b, 110 native English speakers from the U.S. and Canada who speak Spanish as a second language participated in an online study. Participants (70 % female, $M_{\text{age of acquisition}} = 12.42$) needed to meet the same language fluency and eligibility criteria as in the previous studies. The mean English proficiency score was 6.98 ($SD = 0.11$), and the mean Spanish proficiency score was 5.17 ($SD = 1.00$). Two participants were replaced due to having participated in prior studies on the same topic.

4.2. Procedure

Study 3 follows the same procedure as Studies 1, 2a, and 2b. Participants received six pairs of statements as in Study 1. In each pair, one statement was embarrassing in the form of disgust, and one statement was neutral. The disgust set referred to bodily functions and habits, body parts they like to scratch, the shape of their excrement, and so on. The word length of the embarrassing and neutral statements was matched in word count in English ($M_{\text{embarrassing}} = 10.91$, $M_{\text{neutral}} = 10.91$) and Spanish ($M_{\text{embarrassing}} = 11.73$, $M_{\text{neutral}} = 11.27$). Participants chose which statements to say in their native tongue and which to say in the foreign language for each pair. Unlike Studies 1, 2a, and 2b, participants received complete sentences instead of fill-in-the-blank sentences. As with the previous studies, to make sure participants anticipated speaking to an audience, they were told to first make their language choices in an online survey and then participate in a video conference at a later date, where they would read aloud the statements they previously selected in front of the experimenter and other participants.

After making the language choices for all pairs, participants described the criteria that informed their choices in a free response format. The explanations were coded by two native English-speaking researchers blind to the study goal. Participants then rated the emotional and social consequences they anticipated if they were to read aloud the embarrassing statements in English and Spanish, using the same measures from the previous studies. Finally, participants reported their expectations about self-disclosure among native Spanish and native English speakers and indicated whether they felt more emotional and more like themselves when speaking in English or Spanish, using the same scales from Study 1.

Measure	Disgust			Intrusive
	Native	Foreign	Compare	Native	Foreign	Compare
Discomfort	4.84 (2.00)	3.70 (1.87)	$t(220) = 9.83, p < .001, d = 0.66$	4.28 (2.00)	3.43 (1.79)	$t(220) = 7.54, p < .001, d = 0.51$
Negative Social Judgment	4.57 (1.87)	3.93 (1.67)	$t(220) = 6.59, p < .001, d = 0.44$	4.64 (1.90)	4.01 (1.77)	$t(220) = 6.01, p < .001, d = 0.40$
Reveal Kind of Person	3.82 (1.92)	3.34 (1.69)	$t(220) = 4.82, p < .001, d = 0.32$	4.06 (2.00)	3.55 (1.80)	$t(220) = 5.16, p < .001, d = 0.52$
Reveal Personality	3.87 (2.02)	3.10 (1.70)	$t(220) = 6.61, p < .001, d = 0.45$	4.10 (2.00)	3.38 (1.77)	$t(220) = 6.42, p < .001, d = 0.43$

Measure	Native	Foreign	Compare
Discomfort	5.43 (1.67)	4.17 (1.59)	$t(109) = 7.31, p < .001, d = 0.70$
Negative Social Judgment	5.25 (1.91)	4.34 (1.77)	$t(109) = 6.38, p < .001, d = 0.61$
Reveal Kind of Person	4.53 (1.65)	3.70 (1.68)	$t(109) = 5.25, p < .001, d = 0.50$
Reveal Personality	4.15 (2.00)	3.31 (1.67)	$t(109) = 6.21, p < .001, d = 0.59$

Importantly, to measure exposure to or familiarity with each statement, participants reported at the end of the study whether they had experience with each statement (“Has the following ever been true for you?”; Yes, No).

4.3. Results and discussion

Participants preferred to use their foreign language to speak about embarrassing statements. Native English speakers chose to read aloud the embarrassing statements in 68 % of the disgust statement pairs in Spanish, which was significantly higher than chance $Z = -5.89$, $p < .001$, $r = 0.60$. The similarities across all our experiments are striking, given that they involved different languages and embarrassing statements. Fig. 1 shows participants’ language choices in the studies that share the same design (Studies 1, 2a, 2b, and 3).

As in Studies 1, 2a, and 2b, participants expected that using a foreign language would increase emotional and interpersonal distance. Participants reported they felt more like themselves in English (M = 5.9 on a scale from 1 = Spanish to 7 = English; one sample t(109) = 15.51, p < .001, $d_{\text{Cohen}} = 1.48$), and more emotional in English (M = 4.7 on a scale from 1 = Spanish to 7 = English; t(109) = 3.92, p < .001, $d_{\text{Cohen}} = 0.37$) than in Spanish. Replicating Studies 2a and 2b, participants also expected that saying embarrassing statements in their foreign language would carry less serious consequences (Table 8). For example, compared to using their native tongue, participants anticipated that using a foreign language would be less uncomfortable, generate less negative social judgment, and be less reflective of who they were and their true personality.

Participants expected that native Spanish speakers are more likely to be brief when talking about their feelings and reveal private information without hesitation (Table 9). However, only cultural expectations about expressing personal views or opinions correlated with participants’ language choices. Participants expected native Spanish speakers to express their opinions more compared to native English speakers, and this was associated with choosing more embarrassing items to say aloud in Spanish. Thus, cultural expectations of self-disclosure of native English and Spanish speakers may factor into participants’ language choices.

Participants were more likely to choose a foreign language for embarrassing statements due to anticipating more severe emotional and social consequences in their native language. Language choice for disgusting statements most strongly correlated with anticipated discomfort (r = 0.39) and negative social judgments (r = 0.31). This parallels Studies 2a and 2b, suggesting that language choice is associated with the emotional and interpersonal distance a foreign language is expected to provide for aversive statements (see Section 7 in the Supplementary Materials for further analyses of the measures of anticipated consequences).

The qualitative analyses show that participants chose a foreign language for the embarrassing statements, not because of concerns about limited vocabulary or language proficiency. Instead, concerns about aversive emotions were top of mind and guided their language choices. The majority of participants (73.6 %) referred to concepts related to embarrassment or negative emotion as having guided their language choices (Cohen's $\kappa = 0.81$, $z = 8.61$, $p < .001$), while the minority of participants (42.7 %) referenced concepts related to vocabulary (Cohen's $\kappa = 0.70$, $z = 7.4$, $p < .001$). Participants who reported using embarrassment as a determinant of language choice were more likely to choose Spanish, the foreign language, for the embarrassing statements (71.9 %) compared to those who did not mention embarrassment (56.6 %), $\beta = 0.98$, $SE = 0.35$, $\chi^2(1) = 7.83$, $p < .01$. Additionally, participants who mentioned using vocabulary as a determinant of language choice were less likely to choose Spanish for the embarrassing statements (61.3 %) than those who did not mention vocabulary (72.8 %), $\beta = -0.72$, $SE = 0.32$, $\chi^2(1) = 5.12$, $p = .02$, suggesting that vocabulary concerns led people to prefer using their native language to talk about embarrassing statements and thus could not have contributed to the observed pattern of foreign language preference.

Measure	Native English Speakers	Native Spanish Speakers	Compare	Correlations with language Choice
Brevity	3.51 (1.48)	4.15 (1.52)	$t(109) = -3.16, p = .002, d = 0.30$	$r(108) = 0.002, p = .98$
Revealing	3.2 (1.64)	3.73 (1.49)	$t(109) = -2.56, p = .01, d = 0.24$	$r(108) = -0.06, p = .52$
Opinions	4.93 (1.46)	5.08 (1.34)	$t(109) = -0.92, p = .36, d = 0.09$	$r(108) = -0.19, p = .046$

Study 3's goal was to address the possibility that participants might have had limited experience with certain statements, which led them to select the foreign language. The results do not support this idea. Commonness ratings of embarrassing items ($M = 0.71$, $SD = 0.45$) were not significantly different from the ratings of the neutral items ($M = 0.83$, $SD = 0.38$), $t(10) = 1.04$, $p = .33$, $d_{\text{Cohen}} = 0.72$. Thus, participants had experience with the taboo and neutral statements to a similar extent. Moreover, participants' experiences with statements did not correlate with their language choice ($r = -0.32$, $p = .31$).

Participants chose their foreign language to say taboo statements that were just as familiar to them as neutral statements. Combined with Studies 1, 2a, and 2b, Study 3 suggests that people's preferences are related to a desire to emotionally and socially distance themselves from embarrassing statements and cannot be attributed to participants being less familiar with embarrassing statements and being more comfortable lying in their foreign language.

5. Studies 4a and 4b

In Studies 1, 2a, 2b, and 3, we evaluated language choice by forcing participants to use one language for an embarrassing statement and the other language for a neutral statement. This may not reflect what people normally do. People are typically free to use any language they know. In fact, they prefer to use their native tongue in general (Birdsong, 2014). In Studies 4a and 4b, we evaluated our hypothesis without forcing the use of any language, therefore mimicking typical language choice more closely.

Studies 4a and 4b also crossed the languages to further evaluate whether cultural factors could lead to the documented language choice. Study 4a examines the phenomenon with native English speakers who speak Spanish as a foreign language, and Study 4b does so with native Spanish speakers who speak English as a foreign language.

We also consider that perceived language difficulties in one's foreign language may play a role in language choice. If certain topics are more difficult to articulate in a foreign language, this may be reflected in language choice. Thus, participants' language preference may stem more from the perceived difficulty of discussing the topic rather than its anticipated embarrassment.

5.1. Method

5.1.1. Participants

A power analysis for testing the difference between two dependent group means using a two-tailed Wilcoxon signed-rank test for matched pairs suggests at least 208 participants are required to achieve a minimum power of 0.80. Aiming for a robust outcome, we preregistered a sample size of 230 participants for Studies 4a and 4b.

In addition to meeting the eligibility criteria, participants in both studies also had to report having access to a working microphone and being confident with reading aloud sentences at the required level of vocabulary in both Spanish and English.

Study 4a (pre-registered). 270 native English speakers who speak Spanish as a foreign language were recruited from Facebook ads targeting users in the United States. Of these, 21 participants failed at least one attention check and were excluded from the analysis as pre-registered. To remain consistent with our pre-registration, our final sample consists of the first 230 eligible participants (39.6 % Female).5 The mean Spanish proficiency score was 4.63 ($SD = 0.98$), and the mean English score was 6.87 ($SD = 0.21$).

Study 4b (pre-registered). 252 native Spanish speakers who speak English as a second language were recruited from Facebook Ads targeting individuals located in Mexico and Spain. 23 participants failed at least one attention check and were excluded from the analysis as pre-registered. 2 participants were removed from the analysis after completing the survey due to failing a bot detection task. The final sample consists of 230 participants (47 % Female). Participants also reported their Spanish regional dialects: 37 % Spain, 28 % Mexican, 18.7 % Central American, 9.6 % Caribbean, and 7.4 % South American. The mean English proficiency score was 4.79 ($SD = 0.89$), and the mean Spanish score was 6.82 ($SD = 0.33$).

5.2. Procedure

Participants were invited to take part in a study exploring situations and decision-making. To induce anticipated embarrassment, participants were informed that they would record themselves saying a series of statements which would be played for an audience of English-Spanish bilinguals. Participants were first asked to review the statements, select their preferred language for each, and then told they would record themselves saying the statements in the language that they selected. After the main task, participants were informed that there would not be a recording session. Instructions were presented in participants' native language, English in Study 4a and Spanish in Study 4b.

Twenty statements such as 'I would/wouldn't like to try a different hairstyle' were presented in both English and Spanish. Each statement included two possible responses (e.g. would/wouldn't), ensuring that all participants had an option to choose from that fit their experience and situation. Participants were asked to first determine the response that applies to them, imagine saying it aloud in preparation for recording, and then choose the language they would like to use. There were no restrictions on the number of statements that could be assigned to each language, so participants could choose to say all statements in one language if they desired. Participants completed two practice rounds before they reviewed the actual items which were presented randomly along with two attention checks.

The twenty statements were evenly divided into embarrassing and neutral statements and presented in a random order. Embarrassing statements were related to sex, sexual preferences, or sexual habits. The neutral statements matched the embarrassing statements in length and were related to hobbies, experiences, and habitual preferences. The main dependent variable was the rate of choosing to use the foreign language for the embarrassing statements compared to the rate of choosing it for the neutral statements. According to our theory that embarrassment guides language choice, participants should choose the foreign language more for embarrassing items than for neutral items.

Afterward, participants described the criteria that informed their choices in a free response format. After data collection, two native English-speaking assistants (Study 4a) and two native Spanish-speaking assistants (Study 4b) coded the explanations. Participants then rated the emotional and social consequences they anticipated if they were to read aloud the embarrassing statements in English and Spanish, using the same four measures from Study 1.

5.3. Results and discussion

Despite being given free choice to select a language, only a few participants chose to use their native language for all 20 topics (Study 4a: 3.5 %, Study 4b: 0.9 %). Participants in Studies 4a and 4b preferred to use their foreign language to speak about embarrassing statements (Fig. 2). In Study 4a, native English speakers chose to use their foreign language to read aloud 56.8 % of the embarrassing statements compared to 46.6 % of the neutral statements, $Z = -4.19$, $p < .001$, $r = -0.28$. In Study 4b, native Spanish speakers chose to use their foreign language to read aloud 52.3 % of the embarrassing statements, but only 44.4 % of the neutral statements, $Z = -2.35$, $p = .02$, $r = -0.15$. Studies 4a and 4b, then, show the same pattern of results even when languages are crossed. This converges and replicates the results of Studies 2a and 2b, suggesting that language choice was a function of the native-ness of the language, not the language itself or the culture associated with it.

In both studies, participants expected that discussing embarrassing statements in the foreign language would carry less serious consequences. In Study 4a, participants anticipated that using a foreign language would be less uncomfortable, generate less negative social judgment, and be less reflective of who they are and their true personality (Table 10). In Study 4b, participants anticipated that using a foreign language would be less reflective of their true personality (Table 11).

As in previous studies, these expectations may contribute to the language preference for taboo statements in Study 4a. Participants were more likely to choose a foreign language for embarrassing statements when they anticipated more severe emotional or social consequences in their native language: language choice for embarrassing statements most strongly correlated with anticipated embarrassment ($r = 0.24$), negative social judgments ($r = 0.30$), and revealing the kind of person they are ($r = 0.22$). Thus, language choice is related to the emotional and interpersonal distance a foreign language provides for aversive statements. The only exception to this pattern was Study 4b, where language choice for taboo statements was not correlated with anticipated consequences. Section 7 in the Supplementary Materials shows further analyses of the measures of anticipated consequences.

Studies 4a and 4b also included a measure of language difficulty designed to gauge how complex statements may be to articulate. For example, it is possible that participants may prefer to use their native language for statements that are more difficult to articulate, and this may be influencing the observed language choice. We found the opposite. Participants rated embarrassing items as more difficult to articulate in their foreign language ($M = 3.93$, $SD = 1.54$) than their native language ($M = 2.79$, $SD = 1.88$) in Study 4a, $t(229) = 8.31$, $p < .001$. In Study 4b, Embarrassing items were also rated as more difficult to articulate in their foreign language ($M = 4.33$, $SD = 1.92$) than their native language ($M = 1.86$, $SD = 1.36$), $t(229) = 14.776$, $p < .001$. Thus, participants' language preferences were not related to differences in perceived language difficulty. In fact, they selected their foreign language to express the embarrassing statements despite finding them more difficult to articulate in that language.

Measure	Native	Foreign	Compare
Embarrassment	4.62 (1.32)	4.10 (1.32)	$t(229) = 6.30, p < .001, d = 0.42$
Negative social judgment	4.65 (1.24)	4.22 (0.98)	$t(229) = 5.38, p < .001, d = 0.35$
Reveal Kind of Person	4.46 (1.27)	4.15 (1.03)	$t(229) = 4.05, p < .001, d = 0.27$
Reveal Personality	4.51 (1.32)	4.05 (1.12)	$t(229) = 5.30, p < .001, d = 0.55$

Measure	Native	Foreign	Compare
Embarrassment	4.36 (1.29)	4.35 (0.96)	$t(229) = -0.04, p = .97, d = 0.003$
Negative social judgment	4.33 (1.20)	4.29 (1.04)	$t(229) = -0.04, p = .68, d = 0.03$
Reveal Kind of Person	3.84 (1.32)	3.74 (1.17)	$t(229) = -1.26, p = .21, d = 0.08$
Reveal Personality	4.00 (1.26)	3.76 (1.15)	$t(229) = -3.20, p = .002, d = 0.21$

As in Studies 2a, 2b, and 3, participants provided written explanations for how they made their language choices. In Study 4a, more participants (13.5 %) referred to concepts related to embarrassment or negative emotion as having guided their language choices (Cohen's $\kappa = 0.81$, $z = 12.3$, $p < .001$), compared to the number of participants (7.8 %) who referenced concepts related to vocabulary (Cohen's $\kappa = 0.66$, $z = 9.96$, $p < .001$). In Study 4b, participants more frequently referred to concepts related to vocabulary (35 %; Cohen's $\kappa = 0.78$, $z = 11.9$, $p < .001$) than concepts related to embarrassment (10 %; Cohen's $\kappa = 0.63$, $z = 9.96$, $p < .001$). As over 81.7 % of the responses for Study 4a and 55.2 % for Study 4b were uninformative for both concepts of embarrassment and vocabulary, we did not conduct additional analysis on the qualitative responses. Examples from Study 4a include: 'I decided to answer in English', 'It reads smoothly and it's not awkward', 'How embarrassing it would make me look.' In Study 4b, participants gave responses such as 'el nivel de lasemia' ('the level of blasphemy'), 'Which was more comfortable with me', and 'En las reuniones familiares, siempre hablamos en español' ('At family gatherings, we always speak Spanish').

We again leveraged the counterbalanced natures of Studies 4a and 4b by collapsing them into a single analysis to further examine the preference for a foreign language for speaking about embarrassing statements. Participants chose a foreign language to read aloud more than half of the embarrassing statements (54.5 %) compared to less than half of the neutral statements (45.5 %), $Z = -4.45$, $p < .001$, $r = -0.21$. Participants indicated that using a foreign language made them feel less embarrassed, generate less negative social judgment, and be less reflective of who they are and their true personalities (Table 12).

Measure	Native	Foreign	Compare
Embarrassment	4.49 (1.31)	4.22 (0.95)	$t(459) = 3.97, p < .001, d = 0.19$
Negative social judgment	4.49 (1.23)	4.25 (1.01)	$t(459) = 3.59, p < .001, d = 0.17$
Reveal Kind of Person	4.15 (1.33)	3.95 (1.12)	$t(459) = 3.72, p < .001, d = 0.17$
Reveal Personality	4.26 (1.31)	3.91 (1.14)	$t(459) = 6.08, p < .001, d = 0.28$

Language choice was most strongly correlated with anticipated negative social judgments, $r(458) = 0.15, p = .001$. Finally, participants felt that the embarrassing statements were more difficult to articulate in their foreign language ($M = 4.13, SD = 1.75$) than in their native language ($M = 2.32, SD = 1.70$), $t(459) = 16.06, p < .001, d = 0.75$. Despite the anticipated articulation difficulty, participants were more likely to use their foreign language for embarrassing statements, $r(458) = -0.10, p = .04$.

A linear mixed-effects model with the aggregated data found that the type of item (embarrassing vs. neutral) predicted participants' choice of using their foreign language ($\beta = 0.73, p < .001$), even when controlling for foreign language proficiency ($\beta = 0.73, p = .30$), study ($\beta = -0.16, p = .01$), and the interaction between item type and foreign language fluency ($\beta = 0.15, p < .001$), as well as participant-level, and item-level random effects. Thus, the main effect of selecting a foreign language is predicted by embarrassing items, even when controlling for the other variables.

6. General discussion

Across all six studies, bilinguals preferred using a foreign language to say embarrassing statements. This language choice was associated with the emotional and interpersonal distance that a foreign language provides, with the exception of Study 4b where we did not find correlations between anticipated emotional consequences and language choice. Overall, participants chose a foreign language, which they expected would carry less serious consequences when speaking about embarrassing statements. They imagined that the experience would feel less uncomfortable in a foreign language, generate less negative social evaluation, and reveal less about themselves. These findings are predicted by our theory that anticipated emotional impact guides language choice. Indeed, in Studies 2a, 2b, and 3, those who stated embarrassment as a criterion for language choice were more likely to select a foreign language for the taboo statements compared to those who did not consider embarrassment when making their choice.

If the choice to use a foreign language for embarrassing statements is motivated by the expectation that it would increase emotional and interpersonal distance, then the strength of the phenomenon may depend on how similar the languages are to each other. In other words, when a second language more closely resembles the native language, it may feel 'less foreign,' and we might expect a smaller effect. We evaluated this possibility with native Cantonese speakers for whom Mandarin is a non-native language. Although the spoken forms of Mandarin and Cantonese are different (Tang & Van Heuven, 2009), their orthography, vocabulary, and grammar are almost identical, so much so that the Chinese Academy of Social Sciences considers them to be dialects of the same language (Huang, Liao, & Hanyu, 1991). Therefore, we would expect that Mandarin would not be as foreign as English, thereby not providing the same level of distancing as English. As a result, the emotional impact of the statements might not influence the choice between Cantonese and Mandarin. Consistent with this idea, we found no impact of the type of statement on language choice in this case (See Section 9 in the Supplementary Materials).

It is also possible that language proficiency could have influenced language choice in that those with lower proficiency would opt for their first language more frequently. To examine this possibility, we correlated language proficiency in a foreign language with the rate of selecting a foreign language for embarrassing statements. Across studies, we did not find consistent support for such an idea (See Section 10 in Supplementary Materials). While foreign language fluency may play a role in people's language choices, the results of Studies 4a and 4b suggest that participants chose to use their foreign language more despite anticipated articulation difficulties. Furthermore, items related to embarrassment predict people's likelihood of choosing to use their foreign language, even when taking into account foreign language proficiency. Taken together, these results suggest that fluency alone does not account for the language choices observed in these studies.

Languages are also associated with specific cultures and could prime cultural mindsets (Lee, Oyserman, & Bond, 2010), which could very well contribute to language choice in many situations. But culture does not provide a satisfying account for our specific results. Firstly, it is unlikely that cultural norms about specific taboos can explain the language choice observed in our studies because bilinguals consistently choose a foreign language to speak about a variety of taboos, many of which are universal. Secondly, and most importantly, in Studies 2a and 2b, as well as Studies 4a and 4b, English-Spanish bilinguals and Spanish-English bilinguals both chose their respective foreign languages to say embarrassing statements. The consistency of the effect across the two experiments that crossed languages provides strong evidence against a cultural explanation. Though culture could influence language choice in some domains or situations, it cannot explain our results.

The language preference in our studies may reflect a strategy of emotional regulation, where participants preferred using a foreign language, which can help down-regulate and distance themselves from anticipated negative emotions (Santiago-Rivera et al., 2009). Indeed, participants in our studies imagined that using a foreign language would feel less uncomfortable. Another interpretation of our findings draws on the concept of 'face(work)' where individuals utilize different communicative behaviors when interacting with others as a form of self-preservation (Goffman, 1955). Shifting to a foreign language may represent a type of face-saving practice, as participants preferred to use a language they expected would elicit less negative social evaluation and self-disclosure to speak about aversive topics. Prior work in moral reasoning suggests that a foreign language may reduce sensitivity to norms, either by weakening the activation of moral and social norms (Geipel, Hadjichristidis, & Surian, 2015) or by dampening sensitivity to both norms and consequences (Białek, Paruzel-Czachura, & Gawronski, 2019). Our findings suggest that this reduced sensitivity may also extend beyond moral reasoning to shape interpersonal interactions. Additional research can help determine the extent to which language preference reflects a proactive form of emotional regulation or a strategy for managing one's social image, or both.

Another direction for future research is to examine how bilingual language choice can be shaped by the expected audience's language background. Although participants in the present studies were informed that they would be heard by bilinguals like them, we did not directly examine how bilinguals' expectations of audience fluency or native language may influence their language choice. Additionally, the nature of the relationship between audience and speaker may play an interesting role, as certain topics may feel more or less acceptable depending on the perceived social closeness or status of the listener. Future research can also investigate bilinguals' language choice in the context of emotions beyond embarrassment. According to the account that foreign languages reduce emotional intensity, bilinguals may also choose to use a foreign language to speak about aversive topics more generally. On the other hand, this effect may only persist for aversive self-conscious emotions (see Moran & Eyal, 2022), as bilinguals' language choice for embarrassing statements in this experiment was driven partly by a reduction in perceived interpersonal consequences. More broadly, it will be interesting for future research to disentangle the role that language preference plays in emotional regulation face management.

Our study presents experimental evidence for the idea that anticipated emotional reaction guides language choice, and the method we used is a strength but also a limitation. The ecological validity of our method is potentially limited because participants indicated their language choices before speaking. Therefore, participants did not naturally select a language as in an unfolding conversation. This afforded experimental control and avoided confounds but sacrificed naturalness.

In a globalized world where millions are using more than one language every day, it becomes increasingly important to understand bilinguals' cognitive and affective experiences. For example, research has uncovered how using a less emotional foreign language attenuates emotion-based cognitive biases (e.g. Keysar, Hayakawa, & An, 2012). Here, we offer a window into a different part of the bilingual experience by demonstrating that language preferences are guided by anticipated emotional impact, and may serve a regulatory function. People generally feel more comfortable speaking in their native language because of proficiency and familiarity (Birdsong, 2014), but foreign languages offer an emotional distance from aversive, embarrassing topics. This expands our understanding of how billions of bilinguals around the world communicate, choose their language, and experience emotion. The diversity of languages as well as geographical locations across three continents represented in our research further demonstrates its relevance to people around the world. These findings, then, offer a richer understanding of bilinguals' emotional lives and the way they communicate.

As the English actress Kristin Scott Thomas once said, 'I really like acting in French. It's actually quite different for me, from acting in English. It's fun acting in a foreign language. You're liberated or freed from preconceptions.' It turns out that this is true for bilinguals even when they are not acting.

Footnotes (5)

1 These three authors contributed equally to this work.

2 We conducted a resampling procedure based on 10,000 iterations of the stimulus to examine whether oversampling was an issue. We find our results remain reliable and yield the same conclusions presented below (M = 23.53, 95 % CI [23.52, 23.53])

3 Even though Spanish and English are official languages of the island, Spanish is the dominant language. Only 5 % of island-based Puerto Ricans speak English at home and 80 % report speaking English less than "very well" (U.S. Census Bureau, 2008).

4 While the English proficiency was higher compared to the English proficiencies for foreign language in our other studies, Spanish proficiency scores were still significantly higher than English proficiency scores, $t(109) = 5.89$, $p < .001$, $d_{\text{Cohen}} = 0.56$.

5 A total of 249 participants were eligible to be included in the analysis. The extra data was collected due to participants entering the study simultaneously. Here, we report the analysis on the pre-registered sample of 230 participants. The analysis with the full 249 sample shows the same pattern of results as with the pre-registered sample (see Supplemental Materials Section 13).